Welcome, AI Enthusiasts!
Google just dropped a multi-agent model that solves Olympiad math but only for paying users. ChatGPT share links started surfacing in search, exposing private chats. And if you’re using AI at work, a new study says it might be costing you more than you think.
📌 In today’s Generative AI Newsletter:
Gemini Deep Think launches with premium reasoning
ChatGPT chats accidentally surface in search
AI at work carries hidden bias penalties
🧠 Google Rolls Out Gemini ‘Deep Think’ to Tackle Complex Reasoning Tasks
Art scenes made by Google’s AI
Google has released Gemini 2.5 Deep Think, a new version of its AI model that reasons through tough questions by spawning multiple agents to test ideas in parallel. Available to Ultra-tier subscribers starting Friday, the model builds on what was previewed at I/O and is now being positioned as Google’s most capable tool for complex problem-solving.
Here’s what’s rolling out:
Parallel reasoning: generates multiple ideas simultaneously, evaluates them, and converges on the best answer using far more compute than standard models.
IMO pedigree: A variation of the model helped Google win a gold medal at the 2025 International Math Olympiad
SOTA benchmark scores: Gemini 2.5 Deep Think scored 34.8% on Humanity’s Last Exam, beating xAI’s 25.4% and OpenAI’s 20.3%, and led LiveCodeBench 6 with 87.6%.
Built-in tools: The model integrates search and code execution natively, enabling longer, more polished outputs across tasks like web design and software debugging.
The IMO variant will go to researchers; broader testing is coming soon via the Gemini API, focused on enterprise and dev workflows. Multi-agent models are quickly becoming the industry’s new frontier, but they’re not cheap to run. With both Google and xAI locking them behind premium paywalls, the future of high-end reasoning may be less about open access and more about who’s willing to fund the thinking.
🔍 ChatGPT Conversations Were Showing Up on Google
Image Credits: OpenAI
For a brief window, private conversations with ChatGPT were showing up in Google search. Shared chats that users thought were just links became discoverable on the open web, including sensitive information and personal details. After the backlash, OpenAI quietly ended the test.
What went public:
Search crawlers picked up ChatGPT links with the “/share” tag if users opted in to make them discoverable
Personal chats were exposed, including job applications, controversial rants, and bizarre microwave instructions
OpenAI said it was a short-lived experiment, aimed at making helpful conversations easier to find
The company ended the test within hours of public reporting, citing the risk of accidental exposure
As people start using these tools for sensitive reasoning, design choices like this can echo louder than expected. Google has long indexed public links from services like Google Drive, but ChatGPT conversations feel different. They’re often raw, unfiltered, and written in a moment of trust.
💼 Research: The Hidden Penalty of Using AI at Work
Illustration by Sandra Navarro/Credit: Harvard Business
Companies are pushing AI tools onto employees, but a new study shows the social costs may be higher than they think. When 1,026 engineers were asked to evaluate code, they gave significantly lower ratings to work they believed was AI-assisted even though the code was identical in every case. Competence scores dropped 9% on average, and the penalties were steepest for women and engineers over 40.
Inside the findings:
Same code, different label: The only variable was whether reviewers thought AI helped write it
Worse outcomes for women and older workers: Bias deepened when AI use was assumed
Real-world confirmation: At a top tech company, AI tool adoption stalled at 41% after a year
Usage gap: Just 31% of female engineers and 39% of engineers over 40 had tried the tool
The study reveals a blind spot in AI strategy. Companies focus on access, not acceptance. No amount of training or rollout budget will matter if using AI quietly signals incompetence. If trust doesn’t scale with the tech, adoption won’t either.
🚀 Boost your business with us. Advertise where 12M+ AI leaders engage
🌟 Sign up for the first AI Hub in the world.
